Fruit Classification using Convolutional Neural Network(CNN)

Myongkyoon Yang1   Seong In Cho1,*   

1Department of Biosystems Engineering and Biomaterials Science, Seoul National University, Seoul, Republic of Korea

Abstract

Deep learning which is a concept based on an artificial neural network overcomes the previous issues such as optimization or local minima problems. CNN (Convolutional Neural Network) which is one of deep learning algorithms could work well in image processing fields such as classification and recognition. The purpose of this study was to classify images of several fruits using an image training model developed from the CNN algorithm. CAFFE (Convolutional Architecture for Fast Feature Embedding) was used for the CNN algorithm that was developed by BVLC (Berkeley Vision and Learning Center). The fruit images for training were obtained from the data (1,000 categories, 1.2 million images) of ImageNet 2012, Google images and our own images. The 1,000 fruit images of 7 categories were tested and classified. The classification time per each image was about 0.25 seconds and the average classification performance was 0.9166 ranged from 0.8056 to 1.0 depending on fruits. The CNN algorithm was good enough to classify the fruit images into the 7 categories within 10% error. This result showed a better performance as compared to a conventional algorithm using PCA, SVM which had about 88.2% classification error. This shows that CNN which is one of deep learning algorithm can be applied in agriculture area.

Figures & Tables

Fig. 1. A Convolutional Neural Network architecture.